Merge branch 'master' of cane:bkaradzic/bgfx

This commit is contained in:
Branimir Karadžić 2015-11-05 21:15:33 -08:00
commit 0184074392
6 changed files with 857 additions and 740 deletions

View file

@ -2259,6 +2259,7 @@ static void AddDrawListToRenderList(ImVector<ImDrawList*>& out_render_list, ImDr
// If this assert triggers because you are drawing lots of stuff manually, A) workaround by calling BeginChild()/EndChild() to put your draw commands in multiple draw lists, B) #define ImDrawIdx to a 'unsigned int' in imconfig.h and render accordingly.
const unsigned long long int max_vtx_idx = (unsigned long long int)1L << (sizeof(ImDrawIdx)*8);
IM_ASSERT((unsigned long long int)draw_list->_VtxCurrentIdx <= max_vtx_idx);
(void)max_vtx_idx;
GImGui->IO.MetricsRenderVertices += draw_list->VtxBuffer.Size;
GImGui->IO.MetricsRenderIndices += draw_list->IdxBuffer.Size;

View file

@ -508,6 +508,8 @@ struct OcornutImguiContext
m_wm->DockWith(w3, w0, ImGuiWM::E_DOCK_ORIENTATION_BOTTOM);
}
#endif // 0
ImGui::NewFrame();
}
void destroy()

View file

@ -867,6 +867,12 @@ namespace bgfx
}
void Frame::blit(uint8_t _id, TextureHandle _dst, uint8_t _dstMip, uint16_t _dstX, uint16_t _dstY, uint16_t _dstZ, TextureHandle _src, uint8_t _srcMip, uint16_t _srcX, uint16_t _srcY, uint16_t _srcZ, uint16_t _width, uint16_t _height, uint16_t _depth)
{
BX_WARN(m_numBlitItems < BGFX_CONFIG_MAX_BLIT_ITEMS
, "Exceed number of available blit items per frame. BGFX_CONFIG_MAX_BLIT_ITEMS is %d. Skipping blit."
, BGFX_CONFIG_MAX_BLIT_ITEMS
);
if (m_numBlitItems < BGFX_CONFIG_MAX_BLIT_ITEMS)
{
uint16_t item = m_numBlitItems++;
@ -890,6 +896,7 @@ namespace bgfx
key.m_item = item;
m_blitKeys[item] = key.encode();
}
}
void Frame::sort()
{

View file

@ -44,7 +44,7 @@
# ifndef BGFX_CONFIG_RENDERER_METAL
# define BGFX_CONFIG_RENDERER_METAL (0 \
|| BX_PLATFORM_IOS \
|| BX_PLATFORM_OSX \
|| (BX_PLATFORM_OSX >= 101100) \
? 1 : 0)
# endif // BGFX_CONFIG_RENDERER_METAL

View file

@ -712,6 +712,29 @@ namespace bgfx { namespace mtl
uint8_t m_num; // number of color handles
};
struct OcclusionQueryMTL
{
OcclusionQueryMTL()
: m_control(BX_COUNTOF(m_query) )
{
}
void postReset();
void preReset();
void begin(RenderCommandEncoder& _rce, Frame* _render, OcclusionQueryHandle _handle);
void end(RenderCommandEncoder& _rce);
void resolve(Frame* _render, bool _wait = false);
struct Query
{
OcclusionQueryHandle m_handle;
};
Buffer m_buffer;
Query m_query[BGFX_CONFIG_MAX_OCCUSION_QUERIES];
bx::RingBufferControl m_control;
};
} /* namespace metal */ } // namespace bgfx
#endif // BGFX_CONFIG_RENDERER_METAL

View file

@ -348,8 +348,8 @@ namespace bgfx { namespace mtl
if (NULL != NSClassFromString(@"CAMetalLayer") )
{
//on iOS we need the layer as CAmetalLayer
#if BX_PLATFORM_IOS
{
CAMetalLayer* metalLayer = (CAMetalLayer*)g_platformData.nwh;
if (NULL == metalLayer
|| ![metalLayer isKindOfClass:NSClassFromString(@"CAMetalLayer")])
@ -357,14 +357,16 @@ namespace bgfx { namespace mtl
BX_WARN(NULL != m_device, "Unable to create Metal device. Please set platform data window to a CAMetalLayer");
return false;
}
m_metalLayer = metalLayer;
#elif BX_PLATFORM_OSX
// create and set metalLayer
NSWindow* nsWindow = (NSWindow*)g_platformData.nwh;
m_metalLayer = metalLayer;
}
#elif BX_PLATFORM_OSX
{
NSWindow* nsWindow = (NSWindow*)g_platformData.nwh;
[nsWindow.contentView setWantsLayer:YES];
m_metalLayer = [CAMetalLayer layer];
[nsWindow.contentView setLayer:m_metalLayer];
}
#endif // BX_PLATFORM_*
m_device = (id<MTLDevice>)g_platformData.context;
@ -401,8 +403,8 @@ namespace bgfx { namespace mtl
m_uniformBufferFragmentOffset = 0;
g_caps.supported |= (0
| BGFX_CAPS_TEXTURE_COMPARE_LEQUAL
| BGFX_CAPS_TEXTURE_3D
| BGFX_CAPS_TEXTURE_COMPARE_LEQUAL
| BGFX_CAPS_INSTANCING
| BGFX_CAPS_VERTEX_ATTRIB_HALF
// | BGFX_CAPS_FRAGMENT_DEPTH
@ -410,6 +412,9 @@ namespace bgfx { namespace mtl
| BGFX_CAPS_COMPUTE
| BGFX_CAPS_INDEX32
| BGFX_CAPS_DRAW_INDIRECT
// | BGFX_CAPS_TEXTURE_BLIT
// | BGFX_CAPS_TEXTURE_READ_BACK
| BGFX_CAPS_OCCLUSION_QUERY
);
g_caps.maxTextureSize = 2048; //ASK: real caps width/height: 4096, but max depth(3D) size is only: 2048
@ -483,11 +488,15 @@ namespace bgfx { namespace mtl
bx::snprintf(s_viewName[ii], BGFX_CONFIG_MAX_VIEW_NAME_RESERVED+1, "%3d ", ii);
}
m_occlusionQuery.preReset();
return true;
}
void shutdown()
{
m_occlusionQuery.postReset();
for (uint32_t ii = 0; ii < BX_COUNTOF(m_shaders); ++ii)
{
m_shaders[ii].destroy();
@ -507,9 +516,10 @@ namespace bgfx { namespace mtl
MTL_RELEASE(m_samplerDescriptor);
MTL_RELEASE(m_backBufferDepth);
#if BX_PLATFORM_IOS
if (BX_ENABLED(BX_PLATFORM_IOS) )
{
MTL_RELEASE(m_backBufferStencil);
#endif // BX_PLATFORM_*
}
MTL_RELEASE(m_uniformBuffer);
MTL_RELEASE(m_commandQueue);
@ -930,19 +940,7 @@ namespace bgfx { namespace mtl
release(m_backBufferDepth);
}
m_backBufferDepth = m_device.newTextureWithDescriptor(m_textureDescriptor);
#if 0
m_textureDescriptor.pixelFormat = MTLPixelFormatStencil8;
if (NULL != m_backBufferStencil)
{
release(m_backBufferStencil);
}
m_backBufferStencil = m_device.newTextureWithDescriptor(m_textureDescriptor);
#else
m_backBufferStencil = m_backBufferDepth;
#endif // 0
bx::HashMurmur2A murmur;
murmur.begin();
@ -964,14 +962,12 @@ namespace bgfx { namespace mtl
void setShaderUniform(uint8_t _flags, uint32_t _loc, const void* _val, uint32_t _numRegs)
{
if (_flags&BGFX_UNIFORM_FRAGMENTBIT)
{
memcpy(&((char*)m_uniformBuffer.contents())[m_uniformBufferFragmentOffset + _loc], _val, _numRegs*16);
}
else
{
memcpy(&((char*)m_uniformBuffer.contents())[m_uniformBufferVertexOffset + _loc], _val, _numRegs*16);
}
uint32_t offset = 0 != (_flags&BGFX_UNIFORM_FRAGMENTBIT)
? m_uniformBufferFragmentOffset
: m_uniformBufferVertexOffset
;
uint8_t* dst = (uint8_t*)m_uniformBuffer.contents();
memcpy(&dst[offset + _loc], _val, _numRegs*16);
}
void setShaderUniform4f(uint8_t _flags, uint32_t _loc, const void* _val, uint32_t _numRegs)
@ -1026,7 +1022,7 @@ namespace bgfx { namespace mtl
switch ( (uint32_t)type)
{
case UniformType::Mat3:
case UniformType::Mat3|BGFX_UNIFORM_FRAGMENTBIT: \
case UniformType::Mat3|BGFX_UNIFORM_FRAGMENTBIT:
{
float* value = (float*)data;
for (uint32_t ii = 0, count = num/3; ii < count; ++ii, loc += 3*16, value += 9)
@ -1066,15 +1062,18 @@ namespace bgfx { namespace mtl
}
}
void setFrameBuffer(mtl::RenderPassDescriptor renderPassDescriptor, FrameBufferHandle _fbh, bool _msaa = true)
void clearQuad(ClearQuad& _clearQuad, const Rect& _rect, const Clear& _clear, const float _palette[][4])
{
BX_UNUSED(_clearQuad, _rect, _clear, _palette);
}
void setFrameBuffer(RenderPassDescriptor renderPassDescriptor, FrameBufferHandle _fbh, bool _msaa = true)
{
if (!isValid(_fbh) )
{
renderPassDescriptor.colorAttachments[0].texture = m_drawable.texture;
renderPassDescriptor.depthAttachment.texture = m_backBufferDepth;
renderPassDescriptor.stencilAttachment.texture = m_backBufferStencil;
//todo: set resolve textures
}
else
{
@ -1086,15 +1085,13 @@ namespace bgfx { namespace mtl
renderPassDescriptor.colorAttachments[ii].texture = texture.m_ptr;
}
if (isValid(frameBuffer.m_depthHandle))
if (isValid(frameBuffer.m_depthHandle) )
{
const TextureMtl& texture = m_textures[frameBuffer.m_depthHandle.idx];
renderPassDescriptor.depthAttachment.texture = texture.m_ptr;
renderPassDescriptor.stencilAttachment.texture = texture.m_ptrStencil;
//TODO: stencilAttachment should be the same if packed/depth stencil format is used
}
//todo: set resolve textures
}
m_fbh = _fbh;
@ -1104,8 +1101,10 @@ namespace bgfx { namespace mtl
void setDepthStencilState(uint64_t _state, uint64_t _stencil = 0)
{
_state &= BGFX_STATE_DEPTH_WRITE|BGFX_STATE_DEPTH_TEST_MASK;
uint32_t fstencil = unpackStencil(0, _stencil);
uint32_t ref = (fstencil&BGFX_STENCIL_FUNC_REF_MASK)>>BGFX_STENCIL_FUNC_REF_SHIFT;
_stencil &= packStencil(~BGFX_STENCIL_FUNC_REF_MASK, BGFX_STENCIL_MASK);
bx::HashMurmur2A murmur;
@ -1170,6 +1169,7 @@ namespace bgfx { namespace mtl
{
_flags &= BGFX_TEXTURE_SAMPLER_BITS_MASK;
SamplerState sampler = m_samplerStateCache.find(_flags);
if (NULL == sampler)
{
@ -1197,6 +1197,12 @@ namespace bgfx { namespace mtl
return sampler;
}
bool isVisible(Frame* _render, OcclusionQueryHandle _handle, bool _visible)
{
m_occlusionQuery.resolve(_render);
return _visible == (0 != _render->m_occlusion[_handle.idx]);
}
uint32_t getBufferWidth()
{
return m_backBufferDepth.width();
@ -1215,6 +1221,8 @@ namespace bgfx { namespace mtl
uint32_t m_backBufferPixelFormatHash;
uint32_t m_maxAnisotropy;
OcclusionQueryMTL m_occlusionQuery;
Buffer m_uniformBuffer; //todo: use a pool of this
uint32_t m_uniformBufferVertexOffset;
uint32_t m_uniformBufferFragmentOffset;
@ -1255,7 +1263,7 @@ namespace bgfx { namespace mtl
id <CAMetalDrawable> m_drawable;
CommandBuffer m_commandBuffer;
RenderCommandEncoder m_renderCommandEncoder;
};
};
static RendererContextMtl* s_renderMtl;
@ -2102,6 +2110,50 @@ namespace bgfx { namespace mtl
return denseIdx;
}
void OcclusionQueryMTL::postReset()
{
MTL_RELEASE(m_buffer);
}
void OcclusionQueryMTL::preReset()
{
m_buffer = s_renderMtl->m_device.newBufferWithLength(BX_COUNTOF(m_query) * 8, 0);
}
void OcclusionQueryMTL::begin(RenderCommandEncoder& _rce, Frame* _render, OcclusionQueryHandle _handle)
{
while (0 == m_control.reserve(1) )
{
resolve(_render, true);
}
Query& query = m_query[m_control.m_current];
query.m_handle = _handle;
uint32_t offset = _handle.idx * 8;
_rce.setVisibilityResultMode(MTLVisibilityResultModeBoolean, offset);
}
void OcclusionQueryMTL::end(RenderCommandEncoder& _rce)
{
Query& query = m_query[m_control.m_current];
uint32_t offset = query.m_handle.idx * 8;
_rce.setVisibilityResultMode(MTLVisibilityResultModeDisabled, offset);
m_control.commit(1);
}
void OcclusionQueryMTL::resolve(Frame* _render, bool _wait)
{
BX_UNUSED(_wait);
while (0 != m_control.available() )
{
Query& query = m_query[m_control.m_read];
uint64_t result = ( (uint64_t*)m_buffer.contents() )[query.m_handle.idx];
_render->m_occlusion[query.m_handle.idx] = 0 < result;
m_control.consume(1);
}
}
void RendererContextMtl::submit(Frame* _render, ClearQuad& _clearQuad, TextVideoMemBlitter& _textVideoMemBlitter) BX_OVERRIDE
{
BX_UNUSED(_clearQuad);
@ -2179,7 +2231,7 @@ namespace bgfx { namespace mtl
PrimInfo prim = s_primInfo[primIndex];
ProgramMtl* currentProgram = NULL;
mtl::RenderCommandEncoder rce;
RenderCommandEncoder rce;
bool wasCompute = false;
bool viewHasScissor = false;
@ -2193,6 +2245,8 @@ namespace bgfx { namespace mtl
uint32_t statsNumIndices = 0;
uint32_t statsKeyType[2] = {};
m_occlusionQuery.resolve(_render);
if (0 == (_render->m_debug&BGFX_DEBUG_IFH) )
{
bool viewRestart = false;
@ -2227,9 +2281,9 @@ namespace bgfx { namespace mtl
view = key.m_view;
programIdx = invalidHandle;
viewRestart = ( (BGFX_VIEW_STEREO == (_render->m_viewFlags[view] & BGFX_VIEW_STEREO) ) );
viewRestart &= hmdEnabled;
if (viewRestart)
{
if (0 == restartState)
@ -2247,6 +2301,7 @@ namespace bgfx { namespace mtl
}
viewState.m_rect = _render->m_rect[view];
if (viewRestart)
{
viewState.m_rect.m_x = eye * (viewState.m_rect.m_width+1)/2;
@ -2258,7 +2313,8 @@ namespace bgfx { namespace mtl
viewScissorRect = viewHasScissor ? scissorRect : viewState.m_rect;
Clear& clr = _render->m_clear[view];
mtl::RenderPassDescriptor renderPassDescriptor = newRenderPassDescriptor();
RenderPassDescriptor renderPassDescriptor = newRenderPassDescriptor();
renderPassDescriptor.visibilityResultBuffer = m_occlusionQuery.m_buffer;
//todo: check FB size
uint32_t width = getBufferWidth();
@ -2272,9 +2328,10 @@ namespace bgfx { namespace mtl
setFrameBuffer(renderPassDescriptor, fbh);
RenderPassColorAttachmentDescriptor colorAttachment0 = renderPassDescriptor.colorAttachments[0];
if (BGFX_CLEAR_COLOR & clr.m_flags)
if (0 != (BGFX_CLEAR_COLOR & clr.m_flags) )
{
if (BGFX_CLEAR_COLOR_USE_PALETTE & clr.m_flags)
if (0 != (BGFX_CLEAR_COLOR_USE_PALETTE & clr.m_flags) )
{
uint8_t index = (uint8_t)bx::uint32_min(BGFX_CONFIG_MAX_COLOR_PALETTE-1, clr.m_index[0]);
const float* rgba = _render->m_colorPalette[index];
@ -2282,7 +2339,7 @@ namespace bgfx { namespace mtl
const float gg = rgba[1];
const float bb = rgba[2];
const float aa = rgba[3];
colorAttachment0.clearColor = MTLClearColorMake(rr,gg,bb,aa);
colorAttachment0.clearColor = MTLClearColorMake(rr, gg, bb, aa);
}
else
{
@ -2290,20 +2347,25 @@ namespace bgfx { namespace mtl
float gg = clr.m_index[1]*1.0f/255.0f;
float bb = clr.m_index[2]*1.0f/255.0f;
float aa = clr.m_index[3]*1.0f/255.0f;
colorAttachment0.clearColor = MTLClearColorMake(rr,gg,bb,aa);
colorAttachment0.clearColor = MTLClearColorMake(rr, gg, bb, aa);
}
colorAttachment0.loadAction = MTLLoadActionClear;
}
else
{
colorAttachment0.loadAction = MTLLoadActionLoad;
}
//TODO: optimize store actions use discard flag
RenderPassDepthAttachmentDescriptor depthAttachment = renderPassDescriptor.depthAttachment;
if (NULL != depthAttachment.texture)
{
depthAttachment.clearDepth = clr.m_depth;
depthAttachment.loadAction = (BGFX_CLEAR_DEPTH & clr.m_flags) ? MTLLoadActionClear : MTLLoadActionLoad;
depthAttachment.loadAction = 0 != (BGFX_CLEAR_DEPTH & clr.m_flags)
? MTLLoadActionClear
: MTLLoadActionLoad
;
depthAttachment.storeAction = MTLStoreActionStore;
}
@ -2311,7 +2373,10 @@ namespace bgfx { namespace mtl
if (NULL != stencilAttachment.texture)
{
stencilAttachment.clearStencil = clr.m_stencil;
stencilAttachment.loadAction = (BGFX_CLEAR_STENCIL & clr.m_flags) ? MTLLoadActionClear : MTLLoadActionLoad;
stencilAttachment.loadAction = 0 != (BGFX_CLEAR_STENCIL & clr.m_flags)
? MTLLoadActionClear
: MTLLoadActionLoad
;
stencilAttachment.storeAction = MTLStoreActionStore;
}
@ -2319,6 +2384,7 @@ namespace bgfx { namespace mtl
{
m_renderCommandEncoder.endEncoding();
}
rce = m_commandBuffer.renderCommandEncoderWithDescriptor(renderPassDescriptor);
m_renderCommandEncoder = rce;
MTL_RELEASE(renderPassDescriptor);
@ -2332,8 +2398,10 @@ namespace bgfx { namespace mtl
if (BX_ENABLED(BGFX_CONFIG_DEBUG_MTL) )
{
if (item != 1) //ASK: better check ? I don't get the whole restart thing
if (item != 1)
{
rce.popDebugGroup();
}
rce.pushDebugGroup(s_viewName[view]);
}
@ -2347,14 +2415,13 @@ namespace bgfx { namespace mtl
vp.zfar = 1.0f;
rce.setViewport(vp);
if (BGFX_CLEAR_NONE != (clr.m_flags & BGFX_CLEAR_MASK) && !fullscreenRect)
{ //TODO: fallback to clear with quad
//clearQuad(_clearQuad, viewState.m_rect, clr, _render->m_colorPalette);
if (BGFX_CLEAR_NONE != (clr.m_flags & BGFX_CLEAR_MASK)
&& !fullscreenRect)
{
clearQuad(_clearQuad, viewState.m_rect, clr, _render->m_colorPalette);
}
}
//TODO: iscompute
bool resetState = viewChanged || wasCompute;
if (wasCompute)
@ -2370,6 +2437,14 @@ namespace bgfx { namespace mtl
const RenderDraw& draw = renderItem.draw;
const bool hasOcclusionQuery = 0 != (draw.m_stateFlags & BGFX_STATE_INTERNAL_OCCLUSION_QUERY);
if (isValid(draw.m_occlusionQuery)
&& !hasOcclusionQuery
&& !isVisible(_render, draw.m_occlusionQuery, 0 != (draw.m_submitFlags&BGFX_SUBMIT_INTERNAL_OCCLUSION_VISIBLE) ) )
{
continue;
}
const uint64_t newFlags = draw.m_stateFlags;
uint64_t changedFlags = currentState.m_stateFlags ^ draw.m_stateFlags;
currentState.m_stateFlags = newFlags;
@ -2484,12 +2559,12 @@ namespace bgfx { namespace mtl
bool constantsChanged = draw.m_constBegin < draw.m_constEnd;
rendererUpdateUniforms(this, _render->m_uniformBuffer, draw.m_constBegin, draw.m_constEnd);
if (key.m_program != programIdx ||
(BGFX_STATE_BLEND_MASK|BGFX_STATE_BLEND_EQUATION_MASK|BGFX_STATE_ALPHA_WRITE|BGFX_STATE_RGB_WRITE|BGFX_STATE_BLEND_INDEPENDENT|BGFX_STATE_MSAA) & changedFlags ||
currentState.m_vertexBuffer.idx != draw.m_vertexBuffer.idx ||
currentState.m_vertexDecl.idx != draw.m_vertexDecl.idx ||
currentState.m_instanceDataStride != draw.m_instanceDataStride ||
( (blendFactor != draw.m_rgba) && !!(newFlags & BGFX_STATE_BLEND_INDEPENDENT) ) )
if (key.m_program != programIdx
|| (BGFX_STATE_BLEND_MASK|BGFX_STATE_BLEND_EQUATION_MASK|BGFX_STATE_ALPHA_WRITE|BGFX_STATE_RGB_WRITE|BGFX_STATE_BLEND_INDEPENDENT|BGFX_STATE_MSAA) & changedFlags
|| currentState.m_vertexBuffer.idx != draw.m_vertexBuffer.idx
|| currentState.m_vertexDecl.idx != draw.m_vertexDecl.idx
|| currentState.m_instanceDataStride != draw.m_instanceDataStride
|| ( (blendFactor != draw.m_rgba) && !!(newFlags & BGFX_STATE_BLEND_INDEPENDENT) ) )
{
programIdx = key.m_program;
currentState.m_vertexDecl = draw.m_vertexDecl;
@ -2532,14 +2607,14 @@ namespace bgfx { namespace mtl
uint32_t vertexUniformBufferSize = program.m_vshConstantBufferSize;
uint32_t fragmentUniformBufferSize = program.m_fshConstantBufferSize;
if (vertexUniformBufferSize )
if (vertexUniformBufferSize)
{
m_uniformBufferVertexOffset = BX_ALIGN_MASK(m_uniformBufferVertexOffset, program.m_vshConstantBufferAlignmentMask);
rce.setVertexBuffer(m_uniformBuffer, m_uniformBufferVertexOffset, 0);
}
m_uniformBufferFragmentOffset = m_uniformBufferVertexOffset + vertexUniformBufferSize;
if (fragmentUniformBufferSize )
if (fragmentUniformBufferSize)
{
m_uniformBufferFragmentOffset = BX_ALIGN_MASK(m_uniformBufferFragmentOffset, program.m_fshConstantBufferAlignmentMask);
rce.setFragmentBuffer(m_uniformBuffer, m_uniformBufferFragmentOffset, 0);
@ -2591,8 +2666,7 @@ namespace bgfx { namespace mtl
if (currentState.m_vertexBuffer.idx != draw.m_vertexBuffer.idx
|| currentState.m_startVertex != draw.m_startVertex
|| currentState.m_instanceDataBuffer.idx != draw.m_instanceDataBuffer.idx
|| currentState.m_instanceDataOffset != draw.m_instanceDataOffset
)
|| currentState.m_instanceDataOffset != draw.m_instanceDataOffset)
{
currentState.m_vertexBuffer = draw.m_vertexBuffer;
currentState.m_startVertex = draw.m_startVertex;
@ -2635,6 +2709,11 @@ namespace bgfx { namespace mtl
uint32_t numPrimsRendered = 0;
uint32_t numDrawIndirect = 0;
if (hasOcclusionQuery)
{
m_occlusionQuery.begin(rce, _render, draw.m_occlusionQuery);
}
if (isValid(draw.m_indirectBuffer) )
{
// TODO: indirect draw
@ -2677,6 +2756,11 @@ namespace bgfx { namespace mtl
}
}
if (hasOcclusionQuery)
{
m_occlusionQuery.end(rce);
}
statsNumPrimsSubmitted[primIndex] += numPrimsSubmitted;
statsNumPrimsRendered[primIndex] += numPrimsRendered;
statsNumInstances[primIndex] += numInstances;